Inverse response of a RHP zero, calculated
The lab shows the speed falling below zero before it rises. This note follows the first 20 ms in numbers: the voltage the PI asks for, the current jump the zero turns it into, the backward acceleration that follows, and why the dip is a property of the plant that no gain removes.
Companion to the lab experiment The wrong-way start. Run it there; this note predicts what you will see and why.
The model in play
The motor with a right-half-plane zero at z = 5 rad/s and no dead time, PI with kp = 0.15, ki = 0.15 and kd = 0, a 20 rad/s target and no load. The engine multiplies the voltage-to-speed path by (1 − s/z). The voltage is held for 20 ms at a time, so the −u̇/z term acts only at the steps: each change Δu of the held voltage makes the armature current jump by −Δu/(Lz).
Simplifications that matter here
- The current jump is exact for a held voltage; a real amplifier would limit the current, which this model does not.
- The hand estimate ignores back-EMF in the first 20 ms, while the speed is still small.
Predict it by hand
At the first update the whole 20 rad/s is error, so the PI asks for kp·20 + ki·20·Tₛ. The zero turns that step into a current jump the wrong way, and the rotor accelerates backwards at Kₜi/J. The current then relaxes toward u/R with the electrical time constant L/R = 0.06 s.
Check against the engine
The engine's first command is 3.06 V and its current at 0.02 s is −3.21 A, within 0.01 A of the estimate; the difference is the back-EMF of the small reverse speed. The speed bottoms out at −5.31 rad/s at 0.10 s, then the run settles at 1.62 s with 4.47% overshoot. The workbench puts the crossover at 2.16 rad/s, where the zero alone costs arctan(2.16/5) = 23.3° of phase. For slow signals (1 − s/z) ≈ e^(−s/z) acts like a 0.2 s delay, which is why a common rule keeps the crossover below z/2 = 2.5 rad/s.
What to change next
- Move the zero out to s = +20 rad/s.
The same 3.06 V step now makes a current jump four times smaller, −1.28 A, so predict a much shallower dip and more margin. The engine dips only 0.53 rad/s, at 0.04 s, with a 74.20° margin, 0.82% overshoot and settling at 1.32 s.
- Lower P to 0.08 to shrink the first step.
The first command falls to 1.66 V, so if the dip scales with it, predict 5.31 × 1.66/3.06 = 2.88 rad/s. The engine dips 2.76 rad/s, but the PI zero ki/kp moves from 1.00 to 1.88 rad/s and the integral now dominates: overshoot rises to 18.85%, settling to 3.26 s, and the margin falls to 47.64° at 1.49 rad/s.
Common mistakes when reading this run
- Raising P to make the start faster. A larger first step means a deeper wrong-way dip: −12.66 rad/s at P 0.30.
- Treating the zero like an unstable pole. The open loop is stable, with poles at −0.91 and −16.01 s⁻¹; the zero limits how fast the loop can be, not whether a gentle loop is stable.
- Hoping a controller can cancel the zero. Cancelling it would need an unstable pole in the controller at +5 rad/s; the dip is the plant's, and only a slower loop or a different actuator changes it.
Where to go next
Run this experiment in the lab, pin a run and try the two changes above.
Lectures
Terms used in this note
Calculator
Step response calculator: Open the plant alone in the step-response calculator: it undershoots 9.51% before rising to 14.29 (rad/s) per volt.